ParConnect reproducibility report

Flick et al. (2015) [1] proposed a novel distributed algorithm for connected component labeling on de Bruijn graphs. As a special activity of the Student Cluster Competition at SC16 (The International Conference for High Performance Computing, Networking, Storage and Analysis), we replicated the experiments in their paper. Though minor differences exist, experiment results show that, two optimizations, “load balancing” and “active partitions only”, are both effective; the algorithm is scalable when processor cores are enough. We conclude that the correctness, performance and scalability of the algorithm are successfully reproduced.